You photograph the plate.
No tagging, no scrolling a database, no "0.7 cups uncooked rice." Just the camera.
Beta, iOS and Android
Wellnote is a photo-first calorie tracker built for the food you actually eat. Nasi padang, pho, char kway teow, gado-gado, your mom's sambal: snap it, review, done.
Most days you eat in five minutes. Logging that meal should not take longer. Here is what happens between snapping the photo and putting your phone back down.
No tagging, no scrolling a database, no "0.7 cups uncooked rice." Just the camera.
Claude, GPT, and Gemini each identify the dish and estimate portions. Wellnote reconciles them.
If it's right (it usually is), one tap. If portions look off, drag a slider. The log learns.
Why we built this
Existing trackers were built by and for people eating Western diets. Their "nasi padang" entry is one person's home estimate from 2014. Their portion sizes assume a teaspoon of sambal is a teaspoon of sambal.
Wellnote starts from the plate, not the database. It looks at what is actually in front of you, identifies the dish in regional context, and gives you a calorie estimate that is honest about its margin of error.
How it works
Wellnote opens straight to the camera. Frame your meal, tap the shutter, and put your phone down. The upload happens in the background, even on a slow connection, because the log lives on your phone first.
Your image goes to Claude, GPT, and Gemini in parallel. Each one identifies the dish and estimates the portion. Wellnote reconciles the three answers into one estimate, and flags any place they disagreed so you know where to look.
The result lands as a card: dish, components, calories, protein, carbs, fat. If the portion looks off, drag a slider. If the dish is wrong, pick from the alternatives the models offered. Your correction trains the recognizer for next time.
What makes Wellnote different
Three deliberate choices that shaped this app, all of them slightly weird if you're coming from a Silicon Valley calorie tracker.
Vision models still disagree on what they see. Instead of pretending one model is right, Wellnote fans out to Claude, GPT, and Gemini and shows you where they agreed and where they didn't. You get a confident answer when the answer is confident.
Wellnote logs to local SQLite before anything else. Sync runs in the background. You can log on a plane, in a cell-tower dead zone, or in the toilet at the airport. No spinners, no waiting, no data lost when the network drops.
Calorie counts from photos are estimates, not measurements. Wellnote shows the range, not just the midpoint. You see "780 to 880 kcal" when that's true, and you see the single number when the models converge. No false precision.
We are inviting testers in waves through TestFlight on iOS and Expo Go on Android. The basic photo log is free, and will stay free after launch.
Questions
Because guessing wrong all day still adds up. Wellnote is not here to police your portions. It's a quick mirror, not a fence. Most testers use it for two weeks to calibrate their sense of a meal, then check in occasionally.
That's the whole reason it exists. Wellnote runs your photo past three vision models and reconciles their guesses against a database tuned for regional dishes. When we get it wrong, you fix it in a tap, and the recognizer learns from your correction.
On your phone first, in a local SQLite database. A copy syncs to your private Wellnote account on Supabase. We do not sell it, share it with insurers or advertisers, or train public models on your meals. You can export or delete everything at any time.
Free during beta. After launch, the basic photo log stays free. Advanced exports, re-running an old photo through newer models, and weekly nutrition reports will be paid. We will say this clearly before charging anyone, and existing beta testers get the first year on us.
Enter your email on the sign-up page. We invite testers in waves: TestFlight for iOS, Expo Go for Android. Both invitations arrive in your inbox within a couple of days. If you've been waiting longer than a week, email us and we'll fix it.
For Indonesian, Vietnamese, Thai, Filipino, Malaysian, and Singaporean cooking, their database is shallow and inconsistently crowdsourced. Half the entries for "nasi padang" are someone's home guess from a decade ago. Wellnote was built to look at the actual plate instead.